The huge growth in online data storage that has come with the
Internet age is changing the nature of backup operations.
A revolution in backup procedures is taking place, according to
Gartner Dataquest chief analyst Carolyn DiCenzo. "Sans, intelligent
storage appliances and new data replication techniques will
transform the way storage backup and recovery is accomplished over
the next three years," she said.
These new technologies are emerging in response to a real and
pressing need. As Peter Coleman, FCIA Europe president, put it:
"Five years ago you might have 100 Mb to back up, which you could
do in three hours; that has become 3-4Gb, but you've still only got
three hours."
You can't put a quart into a pint pot, and in many cases you
haven't even got the option of trying. E-commerce has meant a
steady move to 24x7 operations even for relatively small
organisations.
The result is that the traditional client-server method of backing
up the data created on the network each day during the night is
increasingly no longer possible: there's too much of it, and the
backup process itself degrades the performance of the increasing
number of other overnight processing tasks.
The major development designed to overcome these problems is the
so-called Lan-free backup. The name is a bit of a misnomer, as Mike
Quinn, Quantum ATL European product marketing manager, explained:
"Lan-free backup is a really nice phrase, but at the end of the day
you need a Lan, even if it's a private one. A San is in effect a
Lan just for data. You're freeing up the network you have, but
you're putting in another one."
The key enabler for Lan-free backup is Fibre Channel connectivity.
This was introduced in the early 1990s for improving transfer to
and from disk, both internally within subsystems and externally to
servers. The use of fibre optic cabling had two benefits over the
prevalent copper SCSI cabling; it substantially increased the
maximum distance between peripheral and host (from a few metres to
several kilometres) and it significantly improved
performance.
Towards the end of the '90s this capability became available on
tape drives and libraries as well. It is still not common in the
mid range and below - Overland Data estimates that only 8 per cent
of systems are currently sold with Fibre Channel connectivity - but
it is becoming increasingly so. Attaching a tape drive via a Fibre
Channel link to a server gives you the benefits of performance and
distance, and also allows you to take at least some of the backup
traffic off the Lan.
But most users are going the extra step and installing a storage
area network (San). At its simplest, this means putting in a switch
between the online storage, the tape drive or library and the
application servers that need to use them.
Switches cost increasingly less, and offer a number of benefits. It
is easier to connect and disconnect both storage devices and
servers without disrupting existing work, and it minimises
recabling when new devices are added.
Most users find it relatively easy to see a return on investment
when installing a San for backup, especially when it is done in
conjunction with consolidating storage. A recent MacArthur Stroud
survey showed that just under one fifth of companies surveyed had
installed a San: of these, 86 per cent (more than six out of seven)
were using it for backup.
Users who wish to retain tape libraries with SCSI interfaces can
still take advantage of the benefits of San by installing a
SCSI-to-Fibre Channel bridge. This type of product is offered by
companies such as Chaparral Network Storage. So using a San can
free up the bandwidth of the existing Lan to improve the
performance of client-server applications. It also offers better
performance than traditional SCSI based storage, and permits siting
of storage devices at greater distances from their hosts. But "the
real benefit", said Coleman, "is Lan-free restore". He argues that
when one of several servers on a Lan goes down, restoring it
affects the performance the users of the other servers get. With
Lan-free there is no such effect.
These benefits have stimulated the take-up of Lan-free software
over the past couple of years. Veritas reports it is selling a lot
of its product, Shared Storage Option (SSO), which is an add-on to
either NetBackup or Backup Exec. Overland Data's European marketing
manager Howard Rippiner estimates that around one third of shops
running multiple servers now use Lan-free.
With a Lan-free architecture, however, the server is still involved
in running the software from the likes of Computer Associates,
Legato and Veritas to make the backups occur. This is not so much a
problem if the server being used is dedicated to storage tasks, as
in the case of a network-attached storage (Nas) filer, but can be
very disruptive on servers used for transaction processing or for
other 24x7 client applications.
Two further developments are emerging to address this problem. The
first is the use of snapshots, a technique originally introduced by
StorageTek to the mainframe world in the early 1990s. Here a
point-in-time copy of a database is made by creating pointers to
the whole content of the database at the given moment. Each time
any part of the database is updated, the new datum is written in an
unused part of the disk leaving the original one unchanged.
As a result, it is possible to access the database both as it
currently stands and as it stood at some specified point in the
past. This allows backup systems to back up a "frozen" copy without
disrupting live processing. You can make a snapshot in a couple of
seconds and back it up at your leisure.
The second new development is server-free backup, an application
that has developed since the arrival of the San and which requires
you to have a San. The principle here is that you have a piece of
software incorporating a data mover agent which sits in a device
within the San that is not a server - a bridge, a router, a hub, or
even a tape drive. The backup software sitting in the server issues
a backup request to this device, which then takes over and manages
the backup between disk and tape, reporting back to the server when
it is finished. In the meantime, the server is free to carry on
performing all its other tasks. Restore works in exactly the same
way.
Apart from freeing the server, this technique also improves backup
performance, as it is not being inhibited by any other process.
Some estimates suggest backups take place 50 per cent faster as a
result.
Veritas was one of the first companies to introduce a serverless
backup software product with its ServerFree Backup Option, launched
a year ago under the code name Vertex.